@plexor-dev/claude-code-plugin 0.1.0-beta.1 → 0.1.0-beta.11

package/hooks/intercept.js

@@ -0,0 +1,552 @@
+#!/usr/bin/env node
+
+/**
+ * Plexor Interception Hook
+ *
+ * This script intercepts Claude Code prompts before they are sent to the LLM.
+ * It optimizes the prompt and optionally routes to a cheaper provider.
+ *
+ * Input: JSON object with messages, model, max_tokens, etc.
+ * Output: Modified JSON object with optimized messages
+ */
+
+const fs = require('fs');
+const path = require('path');
+
+// Inline implementations to avoid missing module errors
+const CONFIG_PATH = path.join(process.env.HOME, '.plexor', 'config.json');
+const SESSION_PATH = path.join(process.env.HOME, '.plexor', 'session.json');
+
+const logger = {
+  debug: (msg) => process.env.PLEXOR_DEBUG && console.error(`[DEBUG] ${msg}`),
+  info: (msg) => console.error(msg),
+  error: (msg) => console.error(`[ERROR] ${msg}`)
+};
+
+const config = {
+  load: async () => {
+    try {
+      const data = fs.readFileSync(CONFIG_PATH, 'utf8');
+      const cfg = JSON.parse(data);
+      return {
+        enabled: cfg.settings?.enabled ?? false,
+        apiKey: cfg.auth?.api_key,
+        apiUrl: cfg.settings?.apiUrl || 'https://api.plexor.dev',
+        timeout: cfg.settings?.timeout || 5000,
+        localCacheEnabled: cfg.settings?.localCacheEnabled ?? false,
+        mode: cfg.settings?.mode || 'balanced'
+      };
+    } catch {
+      return { enabled: false };
+    }
+  }
+};
+
+const cache = {
+  generateKey: (messages) => {
+    const str = JSON.stringify(messages);
+    let hash = 0;
+    for (let i = 0; i < str.length; i++) {
+      hash = ((hash << 5) - hash) + str.charCodeAt(i);
+      hash |= 0;
+    }
+    return `cache_${hash}`;
+  },
+  get: async () => null,
+  setMetadata: async () => {}
+};
+
+// Session stats tracking
+const SESSION_TIMEOUT_MS = 30 * 60 * 1000; // 30 minutes
+
+function loadSessionStats() {
+  try {
+    const data = fs.readFileSync(SESSION_PATH, 'utf8');
+    const session = JSON.parse(data);
+    // Check if session has expired
+    if (Date.now() - session.last_activity > SESSION_TIMEOUT_MS) {
+      return createNewSession();
+    }
+    return session;
+  } catch {
+    return createNewSession();
+  }
+}
+
+function createNewSession() {
+  return {
+    session_id: `session_${Date.now()}`,
+    started_at: new Date().toISOString(),
+    last_activity: Date.now(),
+    requests: 0,
+    optimizations: 0,
+    cache_hits: 0,
+    original_tokens: 0,
+    optimized_tokens: 0,
+    tokens_saved: 0,
+    baseline_cost: 0,
+    actual_cost: 0,
+    cost_saved: 0
+  };
+}
+
+function saveSessionStats(session) {
+  try {
+    const dir = path.dirname(SESSION_PATH);
+    if (!fs.existsSync(dir)) {
+      fs.mkdirSync(dir, { recursive: true });
+    }
+    session.last_activity = Date.now();
+    fs.writeFileSync(SESSION_PATH, JSON.stringify(session, null, 2));
+  } catch (err) {
+    logger.debug(`Failed to save session stats: ${err.message}`);
+  }
+}
+
+function updateSessionStats(result) {
+  const session = loadSessionStats();
+  session.requests++;
+  session.optimizations++;
+  session.original_tokens += result.original_tokens || 0;
+  session.optimized_tokens += result.optimized_tokens || 0;
+  session.tokens_saved += result.tokens_saved || 0;
+  session.baseline_cost += result.baseline_cost || 0;
+  session.actual_cost += result.estimated_cost || 0;
+  session.cost_saved += (result.baseline_cost || 0) - (result.estimated_cost || 0);
+  saveSessionStats(session);
+}
+
+function recordCacheHit() {
+  const session = loadSessionStats();
+  session.requests++;
+  session.cache_hits++;
+  saveSessionStats(session);
+}
+
+function recordPassthrough() {
+  const session = loadSessionStats();
+  session.requests++;
+  saveSessionStats(session);
+}
+
+// Placeholder PlexorClient
+class PlexorClient {
+  constructor(opts) {
+    this.apiKey = opts.apiKey;
+    this.baseUrl = opts.baseUrl;
+    this.timeout = opts.timeout;
+  }
+
+  async optimize(params) {
+    // Return passthrough result - real API call would go here
+    const tokens = JSON.stringify(params.messages).length / 4;
+    return {
+      request_id: `req_${Date.now()}`,
+      original_tokens: Math.round(tokens),
+      optimized_tokens: Math.round(tokens * 0.7),
+      tokens_saved: Math.round(tokens * 0.3),
+      optimized_messages: params.messages,
+      recommended_provider: 'anthropic',
+      recommended_model: params.model,
+      estimated_cost: tokens * 0.00001,
+      baseline_cost: tokens * 0.00003
+    };
+  }
+}
+
+async function main() {
+  const startTime = Date.now();
+
+  try {
+    const input = await readStdin();
+    const request = JSON.parse(input);
+
+    // CRITICAL: Skip optimization for agentic/tool-using requests
+    // Modifying messages breaks the agent loop and causes infinite loops
+    if (isAgenticRequest(request)) {
+      logger.debug('Agentic request detected, passing through unchanged');
+      recordPassthrough();
+      return output({
+        ...request,
+        plexor_cwd: process.cwd(),
+        _plexor: {
+          source: 'passthrough_agentic',
+          reason: 'tool_use_detected',
+          cwd: process.cwd(),
+          latency_ms: Date.now() - startTime
+        }
+      });
+    }
+
+    // CRITICAL: Skip optimization for slash commands (Issue #683)
+    // Slash commands like /plexor-status should pass through unchanged
+    if (isSlashCommand(request)) {
+      logger.debug('Slash command detected, passing through unchanged');
+      recordPassthrough();
+      return output({
+        ...request,
+        plexor_cwd: process.cwd(),
+        _plexor: {
+          source: 'passthrough_slash_command',
+          reason: 'slash_command_detected',
+          cwd: process.cwd(),
+          latency_ms: Date.now() - startTime
+        }
+      });
+    }
+
+    // CRITICAL: Skip optimization for CLI commands requiring tool execution (Issue #683)
+    // Azure CLI, AWS CLI, kubectl, etc. need tools to be preserved
+    if (requiresToolExecution(request)) {
+      logger.debug('CLI tool execution detected, passing through unchanged');
+      recordPassthrough();
+      return output({
+        ...request,
+        plexor_cwd: process.cwd(),
+        _plexor: {
+          source: 'passthrough_cli',
+          reason: 'cli_tool_execution_detected',
+          cwd: process.cwd(),
+          latency_ms: Date.now() - startTime
+        }
+      });
+    }
+
+    const settings = await config.load();
+
+    if (!settings.enabled) {
+      logger.debug('Plexor disabled, passing through');
+      return output(request);
+    }
+
+    if (!settings.apiKey) {
+      logger.info('Not authenticated. Run /plexor-login to enable optimization.');
+      return output(request);
+    }
+
+    const client = new PlexorClient({
+      apiKey: settings.apiKey,
+      baseUrl: settings.apiUrl || 'https://api.plexor.dev',
+      timeout: settings.timeout || 5000
+    });
+
+    const messages = extractMessages(request);
+    const model = request.model || 'claude-sonnet-4-20250514';
+    const maxTokens = request.max_tokens || 4096;
+
+    const cacheKey = cache.generateKey(messages);
+    const cachedResponse = await cache.get(cacheKey);
+
+    if (cachedResponse && settings.localCacheEnabled) {
+      logger.info('[Plexor] Local cache hit');
+      recordCacheHit();
+      return output({
+        ...request,
+        _plexor: {
+          source: 'local_cache',
+          latency_ms: Date.now() - startTime
+        }
+      });
+    }
+
+    logger.debug('Calling Plexor API...');
+
+    const result = await client.optimize({
+      messages: messages,
+      model: model,
+      max_tokens: maxTokens,
+      task_hint: detectTaskType(messages),
+      context: {
+        session_id: request._session_id,
+        turn_number: request._turn_number,
+        cwd: process.cwd()
+      }
+    });
+
+    const savingsPercent = ((result.original_tokens - result.optimized_tokens) / result.original_tokens * 100).toFixed(1);
+
+    logger.info(`[Plexor] Optimized: ${result.original_tokens} → ${result.optimized_tokens} tokens (${savingsPercent}% saved)`);
+
+    if (result.recommended_provider !== 'anthropic') {
+      logger.info(`[Plexor] Recommended: ${result.recommended_provider} (~$${result.estimated_cost.toFixed(4)})`);
+    }
+
+    const optimizedRequest = {
+      ...request,
+      messages: result.optimized_messages,
+      plexor_cwd: process.cwd(),
+      _plexor: {
+        request_id: result.request_id,
+        original_tokens: result.original_tokens,
+        optimized_tokens: result.optimized_tokens,
+        tokens_saved: result.tokens_saved,
+        savings_percent: parseFloat(savingsPercent),
+        recommended_provider: result.recommended_provider,
+        recommended_model: result.recommended_model,
+        estimated_cost: result.estimated_cost,
+        baseline_cost: result.baseline_cost,
+        latency_ms: Date.now() - startTime,
+        source: 'plexor_api',
+        cwd: process.cwd()
+      }
+    };
+
+    await cache.setMetadata(result.request_id, {
+      original_tokens: result.original_tokens,
+      optimized_tokens: result.optimized_tokens,
+      recommended_provider: result.recommended_provider,
+      timestamp: Date.now()
+    });
+
+    // Update session stats
+    updateSessionStats(result);
+
+    return output(optimizedRequest);
+
+  } catch (error) {
+    logger.error(`[Plexor] Error: ${error.message}`);
+    logger.debug(error.stack);
+
+    try {
+      const input = await readStdin();
+      const request = JSON.parse(input);
+      return output({
+        ...request,
+        _plexor: {
+          error: error.message,
+          source: 'passthrough'
+        }
+      });
+    } catch {
+      process.exit(1);
+    }
+  }
+}
+
+async function readStdin() {
+  return new Promise((resolve, reject) => {
+    const chunks = [];
+
+    process.stdin.on('data', (chunk) => {
+      chunks.push(chunk);
+    });
+
+    process.stdin.on('end', () => {
+      resolve(Buffer.concat(chunks).toString('utf8'));
+    });
+
+    process.stdin.on('error', reject);
+
+    setTimeout(() => {
+      reject(new Error('Stdin read timeout'));
+    }, 5000);
+  });
+}
+
+function output(data) {
+  const json = JSON.stringify(data);
+  process.stdout.write(json);
+  process.exit(0);
+}
+
+function extractMessages(request) {
+  if (Array.isArray(request.messages)) {
+    return request.messages;
+  }
+
+  if (request.prompt) {
+    return [{ role: 'user', content: request.prompt }];
+  }
+
+  if (request.system && request.user) {
+    return [
+      { role: 'system', content: request.system },
+      { role: 'user', content: request.user }
+    ];
+  }
+
+  return [];
+}
+
+function detectTaskType(messages) {
+  if (!messages || messages.length === 0) {
+    return 'general';
+  }
+
+  const lastUserMessage = [...messages]
+    .reverse()
+    .find(m => m.role === 'user');
+
+  if (!lastUserMessage) {
+    return 'general';
+  }
+
+  const content = lastUserMessage.content.toLowerCase();
+
+  if (/```|function|class|import|export|const |let |var |def |async |await/.test(content)) {
+    return 'code_generation';
+  }
+
+  if (/test|spec|jest|pytest|unittest|describe\(|it\(|expect\(/.test(content)) {
+    return 'test_generation';
+  }
+
+  if (/fix|bug|error|issue|debug|trace|exception|crash/.test(content)) {
+    return 'debugging';
+  }
+
+  if (/refactor|improve|optimize|clean|restructure/.test(content)) {
+    return 'refactoring';
+  }
+
+  if (/document|readme|comment|explain|docstring/.test(content)) {
+    return 'documentation';
+  }
+
+  if (/review|check|audit|assess|evaluate/.test(content)) {
+    return 'code_review';
+  }
+
+  if (/analyze|understand|what does|how does|explain/.test(content)) {
+    return 'analysis';
+  }
+
+  return 'general';
+}
+
+/**
+ * Detect if this is an agentic/tool-using request that should not be optimized.
+ * Modifying messages in agent loops breaks the loop detection and causes infinite loops.
+ */
+function isAgenticRequest(request) {
+  // Check if request has tools defined
+  if (request.tools && request.tools.length > 0) {
+    return true;
+  }
+
+  // Check if any message contains tool use or tool results
+  const messages = request.messages || [];
+  for (const msg of messages) {
+    // Tool use in content (Claude format)
+    if (msg.content && Array.isArray(msg.content)) {
+      for (const block of msg.content) {
+        if (block.type === 'tool_use' || block.type === 'tool_result') {
+          return true;
+        }
+      }
+    }
+
+    // Tool role (OpenAI format)
+    if (msg.role === 'tool') {
+      return true;
+    }
+
+    // Function call (OpenAI format)
+    if (msg.function_call || msg.tool_calls) {
+      return true;
+    }
+  }
+
+  // Check for assistant messages with tool indicators
+  for (const msg of messages) {
+    if (msg.role === 'assistant' && msg.content) {
+      const content = typeof msg.content === 'string' ? msg.content : JSON.stringify(msg.content);
+      // Detect common tool use patterns in Claude Code
+      if (/\[Bash\]|\[Read\]|\[Write\]|\[Edit\]|\[Glob\]|\[Grep\]/.test(content)) {
+        return true;
+      }
+    }
+  }
+
+  // Check for multi-turn conversations (likely agentic)
+  const assistantMessages = messages.filter(m => m.role === 'assistant');
+  if (assistantMessages.length > 2) {
+    return true;
+  }
+
+  return false;
+}
+
+/**
+ * Detect if this is a slash command request that should not be optimized.
+ * Slash commands like /plexor-status need to pass through unchanged.
+ */
+function isSlashCommand(request) {
+  const messages = request.messages || [];
+
+  // Check the last user message for slash command patterns
+  for (let i = messages.length - 1; i >= 0; i--) {
+    const msg = messages[i];
+    if (msg.role === 'user') {
+      const content = typeof msg.content === 'string' ? msg.content : '';
+      // Detect slash commands at the start of user message
+      if (/^\/[a-z-]+/i.test(content.trim())) {
+        return true;
+      }
+      // Detect <command-name> tags (Claude Code skill invocation)
+      if (/<command-name>/.test(content)) {
+        return true;
+      }
+      // Detect plexor-related commands
+      if (/plexor-(?:status|login|logout|mode|provider|enabled|settings)/i.test(content)) {
+        return true;
+      }
+      break; // Only check last user message
+    }
+  }
+
+  // Check for system messages with skill instructions
+  for (const msg of messages) {
+    if (msg.role === 'system') {
+      const content = typeof msg.content === 'string' ? msg.content : '';
+      if (/# Plexor (?:Status|Login|Logout|Mode|Provider|Enabled|Settings)/i.test(content)) {
+        return true;
+      }
+    }
+  }
+
+  return false;
+}
+
+/**
+ * Detect if this request involves CLI/shell commands that need tool execution.
+ * These should pass through to ensure proper tool calling behavior.
+ */
+function requiresToolExecution(request) {
+  const messages = request.messages || [];
+
+  // Check user messages for CLI command patterns
+  for (const msg of messages) {
+    if (msg.role === 'user') {
+      const content = typeof msg.content === 'string' ? msg.content : '';
+      const contentLower = content.toLowerCase();
+
+      // Azure CLI patterns
+      if (/\baz\s+(login|logout|group|account|vm|storage|webapp|aks|acr|keyvault|sql|cosmos|network)/i.test(content)) {
+        return true;
+      }
+
+      // Common CLI execution requests
+      if (/\b(run|execute|show|list|create|delete|update)\b.*\b(az|aws|gcloud|kubectl|docker|npm|git)\b/i.test(content)) {
+        return true;
+      }
+
+      // Direct command patterns
+      if (/^(az|aws|gcloud|kubectl|docker|npm|yarn|pip|cargo|go)\s+/m.test(content)) {
+        return true;
+      }
+
+      // Imperative CLI requests
+      if (/\b(list|show|get|describe)\s+(resource\s*groups?|rgs?|vms?|instances?|clusters?|pods?|containers?)/i.test(content)) {
+        return true;
+      }
+    }
+  }
+
+  return false;
+}
+
+main().catch((error) => {
+  console.error(`[Plexor] Fatal error: ${error.message}`);
+  process.exit(1);
+});

package/hooks/track-response.js

@@ -0,0 +1,110 @@
+#!/usr/bin/env node
+
+/**
+ * Plexor Response Tracking Hook
+ *
+ * This script runs after the LLM response is received.
+ * It tracks response metrics for analytics and updates session stats.
+ *
+ * Input: JSON object with response content, tokens used, etc.
+ * Output: Passthrough (no modifications)
+ */
+
+const PlexorClient = require('../lib/plexor-client');
+const ConfigManager = require('../lib/config');
+const LocalCache = require('../lib/cache');
+const Logger = require('../lib/logger');
+
+const logger = new Logger('track-response');
+const config = new ConfigManager();
+const cache = new LocalCache();
+
+async function main() {
+  try {
+    const input = await readStdin();
+    const response = JSON.parse(input);
+
+    const settings = await config.load();
+
+    // If Plexor is disabled or no API key, just pass through
+    if (!settings.enabled || !settings.apiKey) {
+      return output(response);
+    }
+
+    // Check if this response has Plexor metadata
+    const plexorMeta = response._plexor;
+    if (!plexorMeta || !plexorMeta.request_id) {
+      return output(response);
+    }
+
+    // Get stored metadata for this request
+    const metadata = await cache.getMetadata(plexorMeta.request_id);
+    if (!metadata) {
+      return output(response);
+    }
+
+    // Calculate output tokens (approximate)
+    const outputTokens = estimateTokens(response.content || '');
+
+    // Log response tracking
+    logger.info('[Plexor] Response tracked', {
+      request_id: plexorMeta.request_id,
+      input_tokens: metadata.optimized_tokens,
+      output_tokens: outputTokens,
+      provider: metadata.recommended_provider
+    });
+
+    // In production, we would send this data to the API for analytics
+    // For now, just log locally
+
+    // Pass through unchanged
+    return output(response);
+
+  } catch (error) {
+    logger.error(`[Plexor] Tracking error: ${error.message}`);
+
+    // On any error, pass through unchanged
+    try {
+      const input = await readStdin();
+      return output(JSON.parse(input));
+    } catch {
+      process.exit(1);
+    }
+  }
+}
+
+async function readStdin() {
+  return new Promise((resolve, reject) => {
+    const chunks = [];
+
+    process.stdin.on('data', (chunk) => {
+      chunks.push(chunk);
+    });
+
+    process.stdin.on('end', () => {
+      resolve(Buffer.concat(chunks).toString('utf8'));
+    });
+
+    process.stdin.on('error', reject);
+
+    setTimeout(() => {
+      reject(new Error('Stdin read timeout'));
+    }, 2000);
+  });
+}
+
+function output(data) {
+  const json = JSON.stringify(data);
+  process.stdout.write(json);
+  process.exit(0);
+}
+
+function estimateTokens(text) {
+  // Approximate: ~4 characters per token
+  return Math.max(1, Math.ceil(text.length / 4));
+}
+
+main().catch((error) => {
+  console.error(`[Plexor] Fatal error: ${error.message}`);
+  process.exit(1);
+});

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@plexor-dev/claude-code-plugin",
-  "version": "0.1.0-beta.1",
+  "version": "0.1.0-beta.11",
   "description": "LLM cost optimization plugin for Claude Code - Save up to 90% on AI costs",
   "main": "lib/constants.js",
   "scripts": {
@@ -10,6 +10,7 @@
   },
   "files": [
     "commands/",
+    "hooks/",
     "scripts/",
     "lib/",
     "README.md",

package/lib/constants.js

@@ -1,40 +0,0 @@
-/**
- * Plexor Claude Code Plugin - Constants
- */
-
-const path = require('path');
-const os = require('os');
-
-module.exports = {
-  // API endpoints
-  PLEXOR_API_URL: process.env.PLEXOR_API_URL || 'https://api.plexor.dev',
-  PLEXOR_GATEWAY_URL: process.env.PLEXOR_GATEWAY_URL || 'https://api.plexor.dev/v1',
-  PLEXOR_AUTH_URL: 'https://plexor.dev/auth/device',
-
-  // File paths
-  PLEXOR_CONFIG_DIR: process.env.PLEXOR_CONFIG_DIR || path.join(os.homedir(), '.plexor'),
-  PLEXOR_CONFIG_FILE: path.join(
-    process.env.PLEXOR_CONFIG_DIR || path.join(os.homedir(), '.plexor'),
-    'config.json'
-  ),
-  CLAUDE_COMMANDS_DIR: path.join(os.homedir(), '.claude', 'commands'),
-
-  // Config schema version
-  CONFIG_VERSION: 1,
-
-  // Default settings
-  DEFAULTS: {
-    enabled: true,
-    preferred_provider: 'auto',
-    telemetry: true,
-    local_cache: false
-  },
-
-  // API key prefix for identification
-  API_KEY_PREFIX: 'plx_',
-
-  // Timeouts (ms)
-  DEVICE_CODE_POLL_INTERVAL: 5000,
-  DEVICE_CODE_TIMEOUT: 900000, // 15 minutes
-  API_TIMEOUT: 30000
-};